Credo Flags 'Bubble' Talk as AI Capex Language Splits Filings
Credo Technology became the first company in the group to name market 'bubble' concerns around AI infrastructure spending, as peers diverged between warning on capex sustainability and doubling down on buildout risk.
Credo Technology Group Holding Ltd (CRDO), a connectivity-solutions supplier to hyperscale data centers, told investors that "many market analysts and other stakeholders have voiced growing concern that the technology sector, including AI, is currently in a stock market 'bubble' characterized by extreme valuations and unsustainable stock price growth," adding that if those views prove correct, it "could result in a market correction or downturn" that would hurt the company. No other company in the group used bubble language, but the underlying worry about capex durability showed up elsewhere.
Marvell Technology, Inc. (MRVL), a semiconductor maker serving data center and communications customers, carried the same caution in back-to-back filings five months apart: the current level of capital expenditure on AI infrastructure "may not be sustainable," and demand could shrink if customers adopt cheaper alternatives to today's AI training and inference costs. The repetition of identical language in March and May filings suggests Marvell kept the risk live rather than softening it as the year progressed.
CoreWeave, Inc. (CRWV), a cloud platform built for AI workloads, disclosed that its operations "require substantial and growing capital expenditures" and that it will need additional capital to fund growth, while also flagging that a large share of revenue depends on a limited number of customers. That combination—capital intensity paired with customer concentration—sets up a funding profile that hinges on continued willingness by a small number of buyers to keep spending.
On the buildout side, Corvex, Inc. (MOVE), an AI infrastructure operator formed through a merger, listed GPU, memory, storage, networking, cooling, data center space, and power costs as direct drivers of its funding requirements in both its March and May filings, tying that capacity expansion explicitly to "the combined company's post-merger operating strategy". XMax Inc. (XMAX), an AI hardware and edge-device maker, described long development cycles and significant working capital needs, warning that enterprise customers often demand extended pilots, custom integrations, and security reviews that lengthen sales cycles and concentrate revenue among large accounts. By August, XMax was still emphasizing that its ability to extend research and development into new AI models and features could raise near-term operating expenses without a guaranteed revenue payoff.
Advanced Micro Devices Inc. (AMD), the chipmaker, kept consistent language across its May and August filings, citing "significant demand" for its AI accelerators since 2024 while cautioning that customer access to data center capacity and energy, along with construction delays, could affect the timing of adoption. Digi Power X Inc. (DGXX), an AI and HPC hosting provider, pointed to a different kind of constraint: the July 2025 Executive Order on "Woke AI" directing federal agencies to procure only large language models meeting truth-seeking and ideological-neutrality standards, alongside the White House's AI Action Plan on permitting, grid expansion, and export controls, as factors that could shape compliance requirements for government-facing AI hosting. Digi Power X repeated its dependence on "global demand for AI computing infrastructure" and its ability to raise capital as risk factors in a May exhibit filing.
A separate cluster of companies embedding AI into existing products described a different risk register, focused on model behavior rather than capex. Elastic N.V. (ESTC), a search and data platform vendor, said that as it integrates a growing number of large language models into its products, it faces "compounding uncertainties" from third-party providers changing model outputs, capabilities, and behavior "without advance notice," which it linked to product reliability, customer trust, and regulatory compliance risk. Elastic also flagged competition for AI and machine learning talent and the possibility that agentic AI tools could change how developers build software altogether.
Calix, Inc. (CALX), a networking software and hardware provider, disclosed that it is rolling out "agentic workflows" for its customer success team and platform customers, while warning that AI models relying on "incomplete, inadequate, inaccurate, biased or otherwise poor quality data" could fail to deliver the efficiency gains it is investing toward. Daktronics Inc. (DAKT), a display manufacturer, took a narrower stance, warning that AI systems could unintentionally disclose confidential information used as inputs and could generate "inaccurate, incomplete, or misleading outputs" that expose the company to reputational harm or contractual claims.
Taken together, two distinct fronts are opening in AI disclosure language over the survey window: infrastructure suppliers and operators debating whether current capex levels can hold, with Credo naming the bubble question directly and Marvell repeating its sustainability caveat across two filings, and product companies like Elastic and Calix building out risk language around the third-party models and agentic tools they are now shipping. Whether hyperscaler capex commitments hold through the next filing cycle will determine which of those two risk narratives gets more attention next.